Consumer Credit Model（一） - 《Credit Scoring Card Stuff》

credit scoring, the risk assessment approach has worked so well since 1950s that there has been little reason to analyse it deeply.
extension from acceptance scoring to behavioural scoring allowed other decisions on whether to extend further credit or whether to try and cross-sell another product to the borrower(behavior score)
three major developments:
- how to move from default-based scoring to profit scoring, to develop models/techniques which assess not just default risk of borrowers but also their profitability.
- because of the changes to more private application channels like the telephone and the internet, lenders are able to customize the loan offers they make to prospective borrowers. In the consumer lending content, the obvious question is how lenders should price for the different default risks of the borrowers.
- how the methodologies giving rise to the credit risk assessment of individual consumer loans can be extended to portfolios of such loans.

applicaiton scorecard:

One way of viewing application scoring is to say that it tries to match a photograph of the consumer on application with a photograph of their status after a fixed time period of borrowing.
It is a classification problem which compares static characteristics on application with a static performance indicator at some fixed time period in the future.

behavior scorecard:

In this analogy, behavioural scoring would be considered as trying to match a video clip of the dynamics of consumers’ performances in the recent past with photographs of their status at some future time point.
However, behavioural scoring uses exactly the same approach as application scoring by translating the consumer’s dynamics during the observation period into a set of statistics – average balance, maximum credit payment, number of times over the credit limit – and then matching these statistics with future status; that is, translating the video clip into a set of measurements of behaviour.

three major problems:

instead of building a static classification model, behavioural scoring could have been developed by using the dynamics of consumers’past behaviour to build a dynamic model
of their future behaviour.
behavioural scoring could have changed the risk measure it was estimating.
shifts of lender’s objectives, from just forecast the default risk to other business goals like profitability or market share. These systems will need to connect the video clip of customers’past behaviour (or a snapshot of a new applicant) with the video clip of their future behaviour which will be needed to estimate their profitability.